prikaz prve stranice dokumenta Lokalizacija u konvolucijskim neuronskim mrežama
  Download
PDF 8.94 MB
undergraduate thesis
Lokalizacija u konvolucijskim neuronskim mrežama
Osijek: Josip Juraj Strossmayer University of Osijek, Department of Mathematics, 2021. urn:nbn:hr:126:735080
Pavičić, Josip
Josip Juraj Strossmayer University of Osijek
Department of Mathematics

Institutional repository: Repository of School of Applied Mathematics and Informatics

Cite this document

Pavičić, J. (2021). Lokalizacija u konvolucijskim neuronskim mrežama (Undergraduate thesis). Osijek: Josip Juraj Strossmayer University of Osijek, Department of Mathematics. Retrieved from https://urn.nsk.hr/urn:nbn:hr:126:735080

Pavičić, Josip. "Lokalizacija u konvolucijskim neuronskim mrežama." Undergraduate thesis, Josip Juraj Strossmayer University of Osijek, Department of Mathematics, 2021. https://urn.nsk.hr/urn:nbn:hr:126:735080

Pavičić, Josip. "Lokalizacija u konvolucijskim neuronskim mrežama." Undergraduate thesis, Josip Juraj Strossmayer University of Osijek, Department of Mathematics, 2021. https://urn.nsk.hr/urn:nbn:hr:126:735080

Pavičić, J. (2021). 'Lokalizacija u konvolucijskim neuronskim mrežama', Undergraduate thesis, Josip Juraj Strossmayer University of Osijek, Department of Mathematics, accessed 16 November 2024, https://urn.nsk.hr/urn:nbn:hr:126:735080

Pavičić J. Lokalizacija u konvolucijskim neuronskim mrežama [Undergraduate thesis]. Osijek: Josip Juraj Strossmayer University of Osijek, Department of Mathematics; 2021 [cited 2024 November 16] Available at: https://urn.nsk.hr/urn:nbn:hr:126:735080

J. Pavičić, "Lokalizacija u konvolucijskim neuronskim mrežama", Undergraduate thesis, Josip Juraj Strossmayer University of Osijek, Department of Mathematics, Osijek, 2021. Available at: https://urn.nsk.hr/urn:nbn:hr:126:735080

Metadata
TitleLokalizacija u konvolucijskim neuronskim mrežama
Title (english)Localization in Convolutional Neural Networks
AuthorJosip Pavičić
MentorDomagoj Matijević (mentor)
GranterJosip Juraj Strossmayer University of Osijek
Department of Mathematics
Osijek
Defense date and country2021-09-27, Croatia
Scientific / art field,
discipline and subdiscipline		NATURAL SCIENCES
Mathematics
Mathematical Logic and Accounting
Abstract
U svakom konvolucijskom sloju neuronske mreže očuvaju se spacijalne informacije objekata sa slike pomoću kojih se može napraviti lokalizacija. To ćemo pokazati pomoću CAM (en. class activation maps) i GAP (en. global average pooling). CAM za svaku klasu određuje diskriminativna područja na slici, na temelju koje konvolucijska neuronska mreža određuje kojoj klasi pripada slika. Drugim riječima, konvolucijska neuronska mreža označi dio slike na temelju kojeg je klasificira za unaprijed zadanu ili predviđenu klasu. Konvolucijska neuronska mreža trenirana je na 16380 slika likova iz popularne serije Simpsoni, dimenzija 255x255 podijeljenih u 6 kategorija. Rad ima pripadni praktični projekt u kojem je implementirana ideja rada. Napravljeni su eksperimenti koji su sastavni dio ovog rada.
Abstract (english)
Spatial information about objects from the pictures is stored in every convolutional neural network layer which can aid with localization. This is presented using CAM (class activation maps) and GAP (global average pooling). CAM determines discriminatory areas for every class on the picture, based on which the convolutional neural network determines how the picture is classified. In other words, convolutional neural network designates a part of the picture based on which it then classifies the picture under a preset class. Convolutional neural network training is based on 16380 pictures of characters from the critically acclaimed series “The Simpsons”, measuring 255x255 split into 6 categories. This bachelor’s thesis contains the accompanying practical project where the thesis idea is carried out. The experiments form an integral part of the thesis.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:126:735080
Study programmeTitle: Bachelor of Mathematics and Computer Science
Study programme type: university
Study level: undergraduate
Academic / professional title: sveučilišni/a prvostupnik/prvostupnica (baccalaureus/baccalaurea) matematike i računarstva (sveučilišni/a prvostupnik/prvostupnica (baccalaureus/baccalaurea) matematike i računarstva)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence In copyright icon
Created on2021-11-02 11:47:31